User avatar
Ed Raket
Posts: 164
Joined: Mon Apr 30, 2012 4:28 pm

Benchmark program

Thu Aug 02, 2012 7:41 pm

My Pi runs really great with current settings! tested under x and with several games

Code: Select all

cpu_freq=900
gpu_freq=450
core_freq=300
I just wondered if there is like a bechmark program that runs under Raspbian without x? i read Eben's post about nbench, but i can't apt-get that program...

shalo
Posts: 74
Joined: Tue May 08, 2012 7:25 pm

Re: Benchmark program

Thu Aug 02, 2012 8:52 pm

Ed Raket wrote:My Pi runs really great with current settings! tested under x and with several games

Code: Select all

cpu_freq=900
gpu_freq=450
core_freq=300
Are you sure those are your settings?

User avatar
Ed Raket
Posts: 164
Joined: Mon Apr 30, 2012 4:28 pm

Re: Benchmark program

Thu Aug 02, 2012 9:08 pm

Oh yes, typing:

Code: Select all

less /boot/config.txt
shows:

Code: Select all

arm_freq=900
gpu_freq=450
core_freq=300
And:

Code: Select all

cat /proc/cpuinfo
Shows: 898.66 BogoMIPS

And it runs verry nice indeed! :mrgreen: Why do you ask?

shalo
Posts: 74
Joined: Tue May 08, 2012 7:25 pm

Re: Benchmark program

Thu Aug 02, 2012 9:39 pm

Ed Raket wrote:Oh yes, typing:

Code: Select all

less /boot/config.txt
shows:

Code: Select all

arm_freq=900
gpu_freq=450
core_freq=300
And:

Code: Select all

cat /proc/cpuinfo
Shows: 898.66 BogoMIPS

And it runs verry nice indeed! :mrgreen: Why do you ask?
Well, it's very odd. You have the gpu running faster than the core, possible but let's see what your settings mean:

core of 300 gives a pll of 1200, possibly 600, not 100% certain.

Now then, what does that mean for your gpu of 450. Well, if the pll is 600 that means your gpu is either running half the time at 600 and half the time at 300 for an average of 450mhz. I'd be very surprised if the 3d block will actually work like that with any level of over-voltage.

If we go with 1200 pll that would mean what, it's actually running at 600mhz 1/4 of the time and 400mhz 3/4 of the time?

Honestly, everything suggests the core can be run much faster than the 3d block which has much less headroom and yet your settings are the opposite and don't really make sense to me. Perhaps I have missed something.

User avatar
Ed Raket
Posts: 164
Joined: Mon Apr 30, 2012 4:28 pm

Re: Benchmark program

Thu Aug 02, 2012 10:07 pm

Is'nt the "core" speed something like the "bus" speed? otherwise i could not guess what "arm" and "gpu" speed changes... :shock: maybe there are some commands or a proggie that i can use to shed some light on this matter.... i used these settings becouse i stumbeled upon a post that mentioned to use GPU 1/2 of ARM speed, and CORE 1/3 of ARM speed. so i did ;)

And im not exactly sure of the effect, thats why i would like to benchmark this setup and compare to other settings.

But i must say.... my i5 laptop seems quite slow all of a suddon... :mrgreen:

portets
Posts: 186
Joined: Sat Oct 29, 2011 6:24 am

Re: Benchmark program

Thu Aug 02, 2012 10:40 pm

The gpu should be equal to or half of core speed. Or 1/3 of 2x core speed.
example:
core_freq=420

420 x 2 = 840
840 / 3 = 280

gpu_freq=280

arm_freq is the cpu, core freq is something related to the interaction between the gpu and cpu. Core freq also changes the cpu's l2 cache speed.

Your gpu_freq is really high. Did you test in quake3? The gpu downclocks when it's not being used much, so a demanding 3d game could crash it.
Core freq can usually be upped more than the gpu_freq. I would instead do:
arm_freq=900
core_freq=420
gpu_freq=280

then play around from there.

To use nbench, download http://www.tux.org/~mayer/linux/nbench- ... 2.3.tar.gz click save.
open up your home folder, your download should be there. right click, extract here. open up the new folder just created. At the top, click tools, open in terminal. then type "make" withut quotes. close all applications as to not effect results(except your terminal). make might take a few minutes.
Then type "./nbench" without quotes. copy and paste your results into a text file and compare.

To run without X, just rename the folder "nbench", boot without X, type "cd nbench", type "./nbench", Write down results.

User avatar
Ed Raket
Posts: 164
Joined: Mon Apr 30, 2012 4:28 pm

Re: Benchmark program

Thu Aug 02, 2012 11:11 pm

portets wrote:The gpu should be equal to or half of core speed. Or 1/3 of 2x core speed.
example:
core_freq=420

420 x 2 = 840
840 / 3 = 280

gpu_freq=280

arm_freq is the cpu, core freq is something related to the interaction between the gpu and cpu. Core freq also changes the cpu's l2 cache speed.

Your gpu_freq is really high. Did you test in quake3? The gpu downclocks when it's not being used much, so a demanding 3d game could crash it.
Core freq can usually be upped more than the gpu_freq. I would instead do:
arm_freq=900
core_freq=420
gpu_freq=280

then play around from there.

To use nbench, download http://www.tux.org/~mayer/linux/nbench- ... 2.3.tar.gz click save.
open up your home folder, your download should be there. right click, extract here. open up the new folder just created. At the top, click tools, open in terminal. then type "make" withut quotes. close all applications as to not effect results(except your terminal). make might take a few minutes.
Then type "./nbench" without quotes. copy and paste your results into a text file and compare.

To run without X, just rename the folder "nbench", boot without X, type "cd nbench", type "./nbench", Write down results.
Thanks for this explanation!, ill download "nbench" tomorrow and have a go on the values, ill post back here what is say's ;)

User avatar
Ed Raket
Posts: 164
Joined: Mon Apr 30, 2012 4:28 pm

Re: Benchmark program

Fri Aug 03, 2012 8:41 am

I have compared my settings with those proposed by "Portets" and the second one gains more speed :D


Settings:

Code: Select all

arm_freq=900
gpu_freq=450
core_freq=300

TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 256.96 : 6.59 : 2.16
STRING SORT : 38.145 : 17.04 : 2.64
BITFIELD : 8.0763e+07 : 13.85 : 2.89
FP EMULATION : 51.698 : 24.81 : 5.72
FOURIER : 2445.1 : 2.78 : 1.56
ASSIGNMENT : 3.0166 : 11.48 : 2.98
IDEA : 868.54 : 13.28 : 3.94
HUFFMAN : 533.81 : 14.80 : 4.73
NEURAL NET : 3.7894 : 6.09 : 2.56
LU DECOMPOSITION : 93.08 : 4.82 : 3.48
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX : 13.624
FLOATING-POINT INDEX: 4.338
Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
CPU :
L2 Cache :
OS : Linux 3.1.9+
C compiler : gcc version 4.6.3 (Debian 4.6.3-8+rpi1)
libc : libc-2.13.so
MEMORY INDEX : 2.833
INTEGER INDEX : 3.898
FLOATING-POINT INDEX: 2.406
Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
* Trademarks are property of their respective holder.

------------------------------------------------------

Settings:

Code: Select all

arm_freq=900
gpu_freq=280
core_freq=420

TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 264.16 : 6.77 : 2.22
STRING SORT : 38.204 : 17.07 : 2.64
BITFIELD : 8.1188e+07 : 13.93 : 2.91
FP EMULATION : 51.997 : 24.95 : 5.76
FOURIER : 2566.5 : 2.92 : 1.64
ASSIGNMENT : 3.2258 : 12.27 : 3.18
IDEA : 872.32 : 13.34 : 3.96
HUFFMAN : 537.63 : 14.91 : 4.76
NEURAL NET : 3.92 : 6.30 : 2.65
LU DECOMPOSITION : 103.68 : 5.37 : 3.88
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX : 13.857
FLOATING-POINT INDEX: 4.622
Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
CPU :
L2 Cache :
OS : Linux 3.1.9+
C compiler : gcc version 4.6.3 (Debian 4.6.3-8+rpi1)
libc : libc-2.13.so
MEMORY INDEX : 2.903
INTEGER INDEX : 3.942
FLOATING-POINT INDEX: 2.563
Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
* Trademarks are property of their respective holder.


Sorry for the "wordwrap", but it's kind off readable... ;)

shalo
Posts: 74
Joined: Tue May 08, 2012 7:25 pm

Re: Benchmark program

Fri Aug 03, 2012 12:19 pm

Ed Raket wrote:Is'nt the "core" speed something like the "bus" speed? otherwise i could not guess what "arm" and "gpu" speed changes... :shock: maybe there are some commands or a proggie that i can use to shed some light on this matter.... i used these settings becouse i stumbeled upon a post that mentioned to use GPU 1/2 of ARM speed, and CORE 1/3 of ARM speed. so i did ;)

And im not exactly sure of the effect, thats why i would like to benchmark this setup and compare to other settings.

But i must say.... my i5 laptop seems quite slow all of a suddon... :mrgreen:
I'm not sure where you could have read any of that, but there is lots of advice and explanations of what all the settings do and even newer settings in the overclocking thread in this very forum.

http://www.raspberrypi.org/phpBB3/viewt ... =29&t=6201

You'll also find other suggestions for benchmarks and stress tests.

User avatar
Ed Raket
Posts: 164
Joined: Mon Apr 30, 2012 4:28 pm

Re: Benchmark program

Fri Aug 03, 2012 2:33 pm

shalo wrote:
Ed Raket wrote:Is'nt the "core" speed something like the "bus" speed? otherwise i could not guess what "arm" and "gpu" speed changes... :shock: maybe there are some commands or a proggie that i can use to shed some light on this matter.... i used these settings becouse i stumbeled upon a post that mentioned to use GPU 1/2 of ARM speed, and CORE 1/3 of ARM speed. so i did ;)

And im not exactly sure of the effect, thats why i would like to benchmark this setup and compare to other settings.

But i must say.... my i5 laptop seems quite slow all of a suddon... :mrgreen:
I'm not sure where you could have read any of that, but there is lots of advice and explanations of what all the settings do and even newer settings in the overclocking thread in this very forum.

http://www.raspberrypi.org/phpBB3/viewt ... =29&t=6201

You'll also find other suggestions for benchmarks and stress tests.
Thx Shalo, i will look in to this ;)

jerrylamos
Posts: 27
Joined: Sun Jul 15, 2012 8:33 pm

Re: Benchmark program

Wed Aug 08, 2012 12:28 pm

Followed the directions for nbench above. This set selected from elinux tested configs.
config.txt
arm_freq=900
core_freq=450
sdram_freq=450

BYTEmark* Native Mode Benchmark ver. 2 (10/95)
Index-split by Andrew D. Balsa (11/97)
Linux/Unix* port by Uwe F. Mayer (12/96,11/97)

TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 270.88 : 6.95 : 2.28
STRING SORT : 38.354 : 17.14 : 2.65
BITFIELD : 8.227e+07 : 14.11 : 2.95
FP EMULATION : 52.396 : 25.14 : 5.80
FOURIER : 2625.8 : 2.99 : 1.68
ASSIGNMENT : 3.443 : 13.10 : 3.40
IDEA : 877.54 : 13.42 : 3.99
HUFFMAN : 541.59 : 15.02 : 4.80
NEURAL NET : 3.92 : 6.30 : 2.65
LU DECOMPOSITION : 107.88 : 5.59 : 4.04
==========================ORIGINAL BYTEMARK RESULTS==========================
INTEGER INDEX : 14.114
FLOATING-POINT INDEX: 4.719
Baseline (MSDOS*) : Pentium* 90, 256 KB L2-cache, Watcom* compiler 10.0
==============================LINUX DATA BELOW===============================
CPU :
L2 Cache :
OS : Linux 3.1.9+
C compiler : gcc version 4.6.3 (Debian 4.6.3-8+rpi1)
libc : libc-2.13.so
MEMORY INDEX : 2.984
INTEGER INDEX : 3.988
FLOATING-POINT INDEX: 2.617
Baseline (LINUX) : AMD K6/233*, 512 KB L2-cache, gcc 2.7.2.3, libc-5.4.38
* Trademarks are property of their respective holder.

The performance I see from the Pi is mostly internet browser of course.
The most important is a good community of forums - there are other boards out there with higher price and better specs I've no interest until they can match Pi's forums.

Jerry

Return to “Advanced users”